@workflow/world-vercel

Production workflow backend for Vercel platform deployments.

Integrates with Vercel's infrastructure for storage, queuing, and authentication. Handles workflow persistence and scaling in production environments.

Used by default for deployments on Vercel. Authentication and API endpoints are configured automatically in Vercel deployments.

Connection failures

Backend connection failures and interrupted event streams follow existing retry policies, including failures with unrecognized error codes. Repeated HTTP/2 session failures rebuild the shared events connection pool. Invalid backend URL protocols, embedded credentials, Fetch-blocked ports, and unsupported request headers fail immediately. Interrupted event writes retain their existing in-process retries; caller cancellations are excluded.

See Backend connection failures for retry behavior and diagnostics.

Events channel

Event writes go over a per-run WebSocket (the default WORKFLOW_EVENTS_TRANSPORT; with http, only for the workflows listed in WORKFLOW_EVENTS_TRANSPORT_WS_OVERRIDE_WORKFLOWS) only while that run's channel is open. The queue handler opens it for each delivery, so workflows need nothing extra. Code that writes a run's events outside a delivery (a custom driver, a long-lived process) opens and releases it itself; otherwise those writes go over HTTP:

import { createWorld, openEventsChannel } from '@workflow/world-vercel';

const world = createWorld();
const release = openEventsChannel(runId);
try {
  await world.events.create(runId, event);
} finally {
  release?.();
}

It returns undefined, and writes stay on HTTP, when the transport is disabled or the World cannot hold a socket (a projectConfig World, as the CLI uses).

Stream writer sockets

Stream writes upgrade to a WebSocket by default; set WORKFLOW_STREAMS_TRANSPORT=http to keep them on HTTP. A writer's first two write groups go over HTTP while the socket opens in the background, and a writer whose upgrade the server declines stays on HTTP.

Releasing a writer after its writes drain retires its socket without closing the shared stream. Reacquiring that handle continues over HTTP; a new step's writer can upgrade independently. Released writers do not keep idle WebSockets or reconnect them in the background.

Step-local getWritable() handles release their transport at step completion, not between acquire/write/release cycles within the step. External Run#getWritable() handles release it on the first observed unlock after pending writes drain; subsequent writes through that same handle stay on HTTP. To retain WebSocket transport across an external streaming burst, hold the writer lock until the burst finishes. Call releaseLock() when finished contributing, not close(), unless you intend to close the shared stream.

Custom dispatcher

HTTP requests (including the queue) default to a shared undici RetryAgent that handles connection pooling and retries. Pass a custom dispatcher to override it, for example, to tune undici on newer Node.js runtimes:

import { Agent } from 'undici';
import { createWorld } from '@workflow/world-vercel';
import { setWorld } from '@workflow/core/runtime';

setWorld(createWorld({ dispatcher: new Agent({ connections: 16 }) }));

Caller user agent

Pass a User-Agent header to append a caller-specific product token while preserving the world-vercel token:

import { createWorld } from '@workflow/world-vercel';

const world = createWorld({
  headers: { 'User-Agent': 'my-framework/1.2.3' },
});